SMOTEHashBoost: Ensemble Algorithm for Imbalanced Dataset Pattern Classification

Class imbalance occurs frequently in machine learning, particularly in binary classification tasks where the majority class has a significantly larger number of samples than the minority class. The majority class is often favored by conventional classifiers, which can lead to biases from improper oversampling or subpar performance when detecting instances of the minority class. Consequently, there is growing concern about algorithmic fairness. Thus, the goal is to improve the ability to identify different discriminatory patterns in the data by creating a large number of test cases, which will result in more fair and unbiased model performance. Here, we introduce SMOTEHashBoost, an ensemble learning approach that overcomes class imbalance by combining three core methods: SMOTE oversampling, hash-based undersampling, and AdaBoost boosting. The proposed approach combines weak classifiers to increase the model’s ability to generalize while generating a balanced and representative dataset. The effectiveness of SMOTEHashBoost is demonstrated by improved performance in handling imbalanced datasets, which makes it a dependable solution for skewed class distributions in real-world classification problems.